Large data breaches are rising, says the OAIC

5 min read · 2 November 2026

No rule changed this week, but the OAIC breach figures did

Hello, no Australian rule changed this week, but the OAIC's latest data breach figures and a date five weeks away point the same way: know where client information sits before something goes wrong. Last week's edition is here if you missed it.

The rule that moved this week

No Australian rule, regulator direction or enforcement action touching AI changed in the past week that I could confirm. What did arrive is a set of figures, and a reminder of a date.

Large data breaches are rising

On 26 October the OAIC released its notifiable data breach statistics for January to June 2026. It recorded 667 notifications, up from 662 in the previous period, and malicious or criminal attacks were the leading cause. Health service providers notified the most breaches (149), but business and professional associations (79) and financial services (51) sit close behind, which matters if you are a bookkeeper, accountant, broker or IT provider holding other people's records.

The Commissioner's advice was to practise data minimisation. In the OAIC's words: "Organisations can't lose personal information they don't hold." Whether the notifiable data breach scheme applies to your business depends on whether the Privacy Act applies to it, including how the small business exemption works, so that is the first thing to settle.

What this means for your business: Look for client files, exported spreadsheets and old email attachments you no longer need, and decide who deletes them. Then check whether your team would recognise a breach if one happened. Reference page: Is it a data breach.

Still coming: 10 December 2026

On 30 September the OAIC released a fact sheet, a flowchart and updated guidance on the new transparency obligation in Australian Privacy Principle 1, subsections 7 to 9. From 10 December 2026, an organisation bound by the Australian Privacy Principles must describe certain automated decisions in its privacy policy. The trigger is a computer program that makes a decision, or does a step substantially and directly tied to one, where the decision could reasonably be expected to significantly affect someone's rights or interests and personal information is used to run it.

The policy must name the kinds of personal information the program uses, the kinds of decisions made solely by the program, and the kinds of decisions where a program does a related step. A law firm's reading of the guidance says the scope is broad, covering rule-based tools and spreadsheets as well as AI, and that human review does not by itself take a tool out of scope. It also describes the obligation as transparency only, with no new right for individuals to contest a decision.

What this means for your business: Start a list of every tool that screens, scores, approves, declines, prices or triages people, and ask each supplier how it works. Reference page: The 10 December 2026 automated decision rule.

Tool changes that affect your client data

Copilot Chat can now draw on past chats

A third-party summary of Microsoft Message Center notice MC1158329 reports that Microsoft 365 Copilot Chat can now infer preferences and work context from earlier chats, on top of stored memories and custom instructions. It reports a staged rollout from mid-September to late October 2026, varying by tenant.

The same summary says the enhanced personalisation control sits at tenant level and is on by default, and that memories are stored in a hidden folder in each user's Exchange mailbox. It also says Purview retention policies and labels do not apply to Copilot Memory, and that memory actions do not generate activity log events. I have not confirmed these details against Microsoft's own documentation, so check your Message Center before relying on them.

The risk is quiet carry-over. A staff member who pastes a client's details into one chat may find the tool drawing on them in a later one, and the owner may not know the setting exists. I found no other tool change this week that I could confirm from a primary source.

What this means for your business: Ask whoever manages your Microsoft 365 whether enhanced personalisation is on, then decide whether staff may use memory features with client information. Reference page: Which AI tools can hold client data.

If staff also use free AI tools on their own, read staff using free AI tools.

Scams to brief your staff on

Networks of AI-built fake sites and endorsements

This is not from the past week; it is the most recent ASIC warning I could open, released on 17 August 2026. ASIC says scammers use generative AI to build webs of fake websites, fake news articles, fake reviews and impersonated public figures that support each other. ASIC removed over 19,400 scams in the 2025-26 financial year, up 182% from 6,915 the year before, and says a basic online search is not enough to verify an opportunity.

What this means for your business: Brief your team that nobody moves money or enters credentials because of an ad, video or endorsement. Verify through contact details you find yourself, using the Scamwatch Stop, Check, Protect steps. Reference page: Do I need an AI policy.

Voice clones and changed payment details

Scamwatch's page on how scammers use technology and AI says a few seconds of recorded voice can be enough to clone it, and advises calling back on a real number you already hold. Its guidance for businesses, Protect your small business from scams, adds two points: be suspicious of an expected invoice with altered payee details, and verify unusual transfer requests that claim to come from the owner or senior staff. Caller ID, email addresses and SMS sender names can be spoofed.

What this means for your business: Make one rule this week: any change to bank details, and any urgent transfer request, is confirmed by calling a number already on file, and a second person approves it. Reference page: Is it a data breach.

One thing to decide this week

Who owns the list of decisions your software makes

Before 10 December, someone has to find out where software makes or supports decisions about people using personal information. Think of applicant screening, lead scoring, credit or quote approvals, pricing, request triage and transaction flags. For each one, note who the decision is about, what information goes in, whether the software decides or only supports a step, and what the supplier can tell you.

The decision this week is not what to write in the policy. It is who owns the list, and by what date. Without an owner, the privacy policy update gets left until the week it is due.

What this means for your business: Name one person, set a date before the end of November, and run your two or three most likely tools through the twelve-question check. Reference page: The Limitless AI Framework twelve-question check.

The bigger picture

Hold less, and know what you hold

The three stories above share one thread. Information you hold is information you must protect, describe in a policy, and keep out of the wrong tools. Tools accumulate memory, software accumulates decisions, and scammers use whatever they can find out about you and your clients. The OAIC's data minimisation advice is the only item here that reduces all three risks at once.

None of this tells you whether your business meets a particular obligation. That depends on your facts, and on whether the Privacy Act applies to you. The reference pages set out the rules and link the sources.

What this means for your business: Choose one category of client information, such as old exports, scanned IDs or email attachments, and decide how long you really need it. Reference page: Does the Privacy Act apply.

That is the week. If one thing in here applies to you, the reference pages say which rule it falls under and link the source, so you can act on it or hand it to your adviser.

More on this: the reference pages · the Limitless AI Framework · something else in mind

See you next Monday.
Ashley

General information for Australian small to medium businesses, not legal advice. Sources are linked in the text.